The cycle's account of AI accountability is strengthened by Du Bois's framework in a specific way: it reframes the question of who counts as an expert. The dominant assumption in AI development is that expertise is located in the people who build the systems—the engineers, the researchers, the executives. Second sight suggests the opposite may often be true. A facial-recognition system that misidentifies darker-skinned faces does so consistently; the people who experience that misidentification possess a knowledge of the system's real behaviour that no internal accuracy metric captures. Auditing AI systems through the testimony of affected communities is not a supplement to technical evaluation. It is, in Du Bois's framework, a primary source of knowledge.
Second sight also connects to the cycle's account of the veil: those behind the veil see out more clearly than the dominant world sees in. The asymmetric glimpse produces the faculty. In the algorithmic context, communities subject to predictive policing, automated credit decisions, or content moderation know things about those systems' behaviour that no engineer sitting behind the algorithm can easily know, because the engineer's experience of the system is the experience of someone never flagged, never misrecognised, never denied. The second sight of those who have been denied is the knowledge the field most needs and most consistently fails to draw upon.
Du Bois was careful to note the limits of this resource. Second sight is exhausting. No one should have to acquire it. The goal is not a population of brilliant double-conscious survivors but a world in which the conditions that produce the faculty have been dismantled. The cycle holds this tension: acknowledging second sight as a genuine epistemological resource while refusing to treat the harm that produces it as acceptable collateral.
Du Bois introduced second sight in The Souls of Black Folk (1903) as the companion concept to double consciousness and the veil. Where double consciousness describes the burden of the asymmetric gaze—the experience of seeing oneself through hostile eyes—second sight describes the faculty that the same burden inadvertently produces. Du Bois drew on William James's pragmatist psychology while transforming it into a sociological claim: that social position shapes epistemic access, and that the position of marginalisation, by requiring a comprehensive knowledge of the dominant system, produces a form of understanding that the dominant position cannot generate.
The concept anticipates standpoint epistemology, which was developed by feminist scholars in the 1970s and 1980s to make a similar argument: that knowledge is always produced from a position, and that marginalised positions have structural access to certain kinds of knowledge that dominant positions cannot reach. Du Bois's version is distinctive in its insistence on the involuntary character of the knowledge—second sight is not chosen; it is required for survival—and in his parallel insistence that the faculty is purchased at a cost that cannot be romanticised away.
Structural Epistemic Access. Second sight is not a talent or a gift in the ordinary sense. It is the structural consequence of occupying a position that requires comprehensive knowledge of the dominant system. The person forced to navigate institutions that were not designed for them develops a knowledge of those institutions' logic and failure modes that the designers rarely possess.
The Knowledge the Builder Lacks. Applied to AI systems: the builder sees the system as designed, the subject sees it as experienced, and the gap is where the harms live. The communities most exposed to algorithmic decision-making possess a second-sight knowledge of how those systems behave under real conditions that no benchmark measures. This knowledge is a primary resource for AI accountability.
Second Sight as Curriculum. Du Bois suggested that the populations living behind the veil understood something not just about the dominant society but about the human condition: that suffering and exclusion produced a moral and spiritual knowledge unavailable from the comfortable centre. Applied to AI: the people first subjected to being reduced to data are the first to learn what survives the reduction—and their knowledge of what it means to remain human under conditions designed to deny it may become, under the pressure of ubiquitous algorithmic assessment, a knowledge that all of humanity will need.
The Limits of the Resource. Second sight is purchased with suffering. No account of its epistemic value can justify the conditions that produce it. Du Bois never romanticised the wound; he wanted to abolish it. The cycle holds the same tension: drawing on the second sight of affected communities while refusing to treat their exposure to harm as an acceptable knowledge-production mechanism.